nonbinary.computer
A better Rust ATProto crate
removed outdated doc
-18
-18
crates/jacquard-repo/tests/interop.rs
-18
crates/jacquard-repo/tests/interop.rs
···
2
2
//!
3
3
//! See: https://github.com/bluesky-social/atproto-interop-tests/tree/main/mst
4
4
//!
5
-
//! ## Current Status (Determinism Bug)
6
-
//!
7
-
//! ### Fixed Issues
8
-
//! 1. **split_around bug**: When split_idx=0 (all entries go right), node.left wasn't being split
9
-
//! - Fixed in tree.rs:640-649 by checking node.left when left_entries is empty
10
-
//! 2. **insert_entry_at_layer bug**: When insert_idx=0 with node.left, left pointer wasn't split
11
-
//! - Fixed in tree.rs:520-565 by splitting node.left around new key
12
-
//!
13
-
//! ### Remaining Issues
14
-
//! - **Non-deterministic tree structure**: Forward vs reverse insertion produces different root CIDs
15
-
//! - All keys are retrievable (no corruption)
16
-
//! - But tree structure differs (different node layouts/pointers)
17
-
//! - Fails even with just 10 keys (test_first_10_keys_determinism)
18
-
//!
19
-
//! ### Next Steps
20
-
//! - Compare tree architecture with rsky-repo and atproto implementations
21
-
//! - May need to restructure how nodes/splits are handled to match reference implementations
22
-
//! - Possible issues: recompress logic, subtree attachment, split handling at different layers
23
5
24
6
use std::sync::Arc;
25
7